IsoBayes: a Bayesian approach for single-isoform proteomics inference

Area of Science:

• Proteomics
• Bioinformatics
• Systems Biology

Background:

• Studying protein isoforms is crucial in biomedical research.
• Current bottom-up mass spectrometry proteomics infers protein isoforms indirectly from peptide identifications, facing challenges with noisy detection and shared peptides.
• Existing methods often abstract results to gene-level or protein isoform groups, hindering isoform-specific analysis.

Purpose of the Study:

• To introduce IsoBayes, a novel statistical method for direct protein isoform-level inference.
• To integrate mass spectrometry proteomics and transcriptomics data within a Bayesian probabilistic framework.
• To enhance the accuracy of protein isoform detection and quantification.

Main Methods:

• Developed a Bayesian probabilistic framework integrating proteomics and transcriptomics data.
• Employed a two-layer latent variable approach to address measurement uncertainty: peptide detection validation and abundance allocation.
• Inferred protein isoform presence/absence (posterior probability) and estimated abundance with credible intervals.

Main Results:

• IsoBayes demonstrated good sensitivity and specificity in detecting protein isoforms through simulations and real datasets.
• Estimated protein isoform abundances showed high correlation with ground truth data.
• The method successfully identified isoforms with significant differences between transcript and protein relative abundances.

Conclusions:

• IsoBayes enables accurate inference at the protein isoform level, overcoming limitations of current proteomics approaches.
• The integration of transcriptomics data enhances the robustness of protein isoform analysis.
• IsoBayes is available as a Bioconductor R package for broader research application.